xref: /freebsd/sys/nlm/nlm_prot_impl.c (revision 724b4bfdf1306e4f2c451b6d146fe0fe0353b2c8)
1 /*-
2  * Copyright (c) 2008 Isilon Inc http://www.isilon.com/
3  * Authors: Doug Rabson <dfr@rabson.org>
4  * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include "opt_inet6.h"
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include <sys/param.h>
34 #include <sys/fail.h>
35 #include <sys/fcntl.h>
36 #include <sys/kernel.h>
37 #include <sys/kthread.h>
38 #include <sys/lockf.h>
39 #include <sys/malloc.h>
40 #include <sys/mount.h>
41 #if __FreeBSD_version >= 700000
42 #include <sys/priv.h>
43 #endif
44 #include <sys/proc.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/syscall.h>
48 #include <sys/sysctl.h>
49 #include <sys/sysent.h>
50 #include <sys/syslog.h>
51 #include <sys/sysproto.h>
52 #include <sys/systm.h>
53 #include <sys/taskqueue.h>
54 #include <sys/unistd.h>
55 #include <sys/vnode.h>
56 
57 #include <nfs/nfsproto.h>
58 #include <nfs/nfs_lock.h>
59 
60 #include <nlm/nlm_prot.h>
61 #include <nlm/sm_inter.h>
62 #include <nlm/nlm.h>
63 #include <rpc/rpc_com.h>
64 #include <rpc/rpcb_prot.h>
65 
66 MALLOC_DEFINE(M_NLM, "NLM", "Network Lock Manager");
67 
68 /*
69  * If a host is inactive (and holds no locks) for this amount of
70  * seconds, we consider it idle and stop tracking it.
71  */
72 #define NLM_IDLE_TIMEOUT	30
73 
74 /*
75  * We check the host list for idle every few seconds.
76  */
77 #define NLM_IDLE_PERIOD		5
78 
79 /*
80  * We only look for GRANTED_RES messages for a little while.
81  */
82 #define NLM_EXPIRE_TIMEOUT	10
83 
84 /*
85  * Support for sysctl vfs.nlm.sysid
86  */
87 static SYSCTL_NODE(_vfs, OID_AUTO, nlm, CTLFLAG_RW, NULL,
88     "Network Lock Manager");
89 static SYSCTL_NODE(_vfs_nlm, OID_AUTO, sysid, CTLFLAG_RW, NULL, "");
90 
91 /*
92  * Syscall hooks
93  */
94 static int nlm_syscall_offset = SYS_nlm_syscall;
95 static struct sysent nlm_syscall_prev_sysent;
96 #if __FreeBSD_version < 700000
97 static struct sysent nlm_syscall_sysent = {
98 	(sizeof(struct nlm_syscall_args) / sizeof(register_t)) | SYF_MPSAFE,
99 	(sy_call_t *) nlm_syscall
100 };
101 #else
102 MAKE_SYSENT(nlm_syscall);
103 #endif
104 static bool_t nlm_syscall_registered = FALSE;
105 
106 /*
107  * Debug level passed in from userland. We also support a sysctl hook
108  * so that it can be changed on a live system.
109  */
110 static int nlm_debug_level;
111 SYSCTL_INT(_debug, OID_AUTO, nlm_debug, CTLFLAG_RW, &nlm_debug_level, 0, "");
112 
113 #define NLM_DEBUG(_level, args...)			\
114 	do {						\
115 		if (nlm_debug_level >= (_level))	\
116 			log(LOG_DEBUG, args);		\
117 	} while(0)
118 #define NLM_ERR(args...)			\
119 	do {					\
120 		log(LOG_ERR, args);		\
121 	} while(0)
122 
123 /*
124  * Grace period handling. The value of nlm_grace_threshold is the
125  * value of time_uptime after which we are serving requests normally.
126  */
127 static time_t nlm_grace_threshold;
128 
129 /*
130  * We check for idle hosts if time_uptime is greater than
131  * nlm_next_idle_check,
132  */
133 static time_t nlm_next_idle_check;
134 
135 /*
136  * A socket to use for RPC - shared by all IPv4 RPC clients.
137  */
138 static struct socket *nlm_socket;
139 
140 #ifdef INET6
141 
142 /*
143  * A socket to use for RPC - shared by all IPv6 RPC clients.
144  */
145 static struct socket *nlm_socket6;
146 
147 #endif
148 
149 /*
150  * An RPC client handle that can be used to communicate with the local
151  * NSM.
152  */
153 static CLIENT *nlm_nsm;
154 
155 /*
156  * An AUTH handle for the server's creds.
157  */
158 static AUTH *nlm_auth;
159 
160 /*
161  * A zero timeval for sending async RPC messages.
162  */
163 struct timeval nlm_zero_tv = { 0, 0 };
164 
165 /*
166  * The local NSM state number
167  */
168 int nlm_nsm_state;
169 
170 
171 /*
172  * A lock to protect the host list and waiting lock list.
173  */
174 static struct mtx nlm_global_lock;
175 
176 /*
177  * Locks:
178  * (l)		locked by nh_lock
179  * (s)		only accessed via server RPC which is single threaded
180  * (g)		locked by nlm_global_lock
181  * (c)		const until freeing
182  * (a)		modified using atomic ops
183  */
184 
185 /*
186  * A pending client-side lock request, stored on the nlm_waiting_locks
187  * list.
188  */
189 struct nlm_waiting_lock {
190 	TAILQ_ENTRY(nlm_waiting_lock) nw_link; /* (g) */
191 	bool_t		nw_waiting;	       /* (g) */
192 	nlm4_lock	nw_lock;	       /* (c) */
193 	union nfsfh	nw_fh;		       /* (c) */
194 	struct vnode	*nw_vp;		       /* (c) */
195 };
196 TAILQ_HEAD(nlm_waiting_lock_list, nlm_waiting_lock);
197 
198 struct nlm_waiting_lock_list nlm_waiting_locks; /* (g) */
199 
200 /*
201  * A pending server-side asynchronous lock request, stored on the
202  * nh_pending list of the NLM host.
203  */
204 struct nlm_async_lock {
205 	TAILQ_ENTRY(nlm_async_lock) af_link; /* (l) host's list of locks */
206 	struct task	af_task;	/* (c) async callback details */
207 	void		*af_cookie;	/* (l) lock manager cancel token */
208 	struct vnode	*af_vp;		/* (l) vnode to lock */
209 	struct flock	af_fl;		/* (c) lock details */
210 	struct nlm_host *af_host;	/* (c) host which is locking */
211 	CLIENT		*af_rpc;	/* (c) rpc client to send message */
212 	nlm4_testargs	af_granted;	/* (c) notification details */
213 	time_t		af_expiretime;	/* (c) notification time */
214 };
215 TAILQ_HEAD(nlm_async_lock_list, nlm_async_lock);
216 
217 /*
218  * NLM host.
219  */
220 enum nlm_host_state {
221 	NLM_UNMONITORED,
222 	NLM_MONITORED,
223 	NLM_MONITOR_FAILED,
224 	NLM_RECOVERING
225 };
226 
227 struct nlm_rpc {
228 	CLIENT		*nr_client;    /* (l) RPC client handle */
229 	time_t		nr_create_time; /* (l) when client was created */
230 };
231 
232 struct nlm_host {
233 	struct mtx	nh_lock;
234 	volatile u_int	nh_refs;       /* (a) reference count */
235 	TAILQ_ENTRY(nlm_host) nh_link; /* (g) global list of hosts */
236 	char		nh_caller_name[MAXNAMELEN]; /* (c) printable name of host */
237 	uint32_t	nh_sysid;	 /* (c) our allocaed system ID */
238 	char		nh_sysid_string[10]; /* (c) string rep. of sysid */
239 	struct sockaddr_storage	nh_addr; /* (s) remote address of host */
240 	struct nlm_rpc	nh_srvrpc;	 /* (l) RPC for server replies */
241 	struct nlm_rpc	nh_clntrpc;	 /* (l) RPC for client requests */
242 	rpcvers_t	nh_vers;	 /* (s) NLM version of host */
243 	int		nh_state;	 /* (s) last seen NSM state of host */
244 	enum nlm_host_state nh_monstate; /* (l) local NSM monitoring state */
245 	time_t		nh_idle_timeout; /* (s) Time at which host is idle */
246 	struct sysctl_ctx_list nh_sysctl; /* (c) vfs.nlm.sysid nodes */
247 	uint32_t	nh_grantcookie;  /* (l) grant cookie counter */
248 	struct nlm_async_lock_list nh_pending; /* (l) pending async locks */
249 	struct nlm_async_lock_list nh_granted; /* (l) granted locks */
250 	struct nlm_async_lock_list nh_finished; /* (l) finished async locks */
251 };
252 TAILQ_HEAD(nlm_host_list, nlm_host);
253 
254 static struct nlm_host_list nlm_hosts; /* (g) */
255 static uint32_t nlm_next_sysid = 1;    /* (g) */
256 
257 static void	nlm_host_unmonitor(struct nlm_host *);
258 
259 struct nlm_grantcookie {
260 	uint32_t	ng_sysid;
261 	uint32_t	ng_cookie;
262 };
263 
264 static inline uint32_t
265 ng_sysid(struct netobj *src)
266 {
267 
268 	return ((struct nlm_grantcookie *)src->n_bytes)->ng_sysid;
269 }
270 
271 static inline uint32_t
272 ng_cookie(struct netobj *src)
273 {
274 
275 	return ((struct nlm_grantcookie *)src->n_bytes)->ng_cookie;
276 }
277 
278 /**********************************************************************/
279 
280 /*
281  * Initialise NLM globals.
282  */
283 static void
284 nlm_init(void *dummy)
285 {
286 	int error;
287 
288 	mtx_init(&nlm_global_lock, "nlm_global_lock", NULL, MTX_DEF);
289 	TAILQ_INIT(&nlm_waiting_locks);
290 	TAILQ_INIT(&nlm_hosts);
291 
292 	error = syscall_register(&nlm_syscall_offset, &nlm_syscall_sysent,
293 	    &nlm_syscall_prev_sysent);
294 	if (error)
295 		NLM_ERR("Can't register NLM syscall\n");
296 	else
297 		nlm_syscall_registered = TRUE;
298 }
299 SYSINIT(nlm_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_init, NULL);
300 
301 static void
302 nlm_uninit(void *dummy)
303 {
304 
305 	if (nlm_syscall_registered)
306 		syscall_deregister(&nlm_syscall_offset,
307 		    &nlm_syscall_prev_sysent);
308 }
309 SYSUNINIT(nlm_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_uninit, NULL);
310 
311 /*
312  * Create a netobj from an arbitrary source.
313  */
314 void
315 nlm_make_netobj(struct netobj *dst, caddr_t src, size_t srcsize,
316     struct malloc_type *type)
317 {
318 
319 	dst->n_len = srcsize;
320 	dst->n_bytes = malloc(srcsize, type, M_WAITOK);
321 	memcpy(dst->n_bytes, src, srcsize);
322 }
323 
324 /*
325  * Copy a struct netobj.
326  */
327 void
328 nlm_copy_netobj(struct netobj *dst, struct netobj *src,
329     struct malloc_type *type)
330 {
331 
332 	nlm_make_netobj(dst, src->n_bytes, src->n_len, type);
333 }
334 
335 
336 /*
337  * Create an RPC client handle for the given (address,prog,vers)
338  * triple using UDP.
339  */
340 static CLIENT *
341 nlm_get_rpc(struct sockaddr *sa, rpcprog_t prog, rpcvers_t vers)
342 {
343 	char *wchan = "nlmrcv";
344 	const char* protofmly;
345 	struct sockaddr_storage ss;
346 	struct socket *so;
347 	CLIENT *rpcb;
348 	struct timeval timo;
349 	RPCB parms;
350 	char *uaddr;
351 	enum clnt_stat stat = RPC_SUCCESS;
352 	int rpcvers = RPCBVERS4;
353 	bool_t do_tcp = FALSE;
354 	bool_t tryagain = FALSE;
355 	struct portmap mapping;
356 	u_short port = 0;
357 
358 	/*
359 	 * First we need to contact the remote RPCBIND service to find
360 	 * the right port.
361 	 */
362 	memcpy(&ss, sa, sa->sa_len);
363 	switch (ss.ss_family) {
364 	case AF_INET:
365 		((struct sockaddr_in *)&ss)->sin_port = htons(111);
366 		protofmly = "inet";
367 		so = nlm_socket;
368 		break;
369 
370 #ifdef INET6
371 	case AF_INET6:
372 		((struct sockaddr_in6 *)&ss)->sin6_port = htons(111);
373 		protofmly = "inet6";
374 		so = nlm_socket6;
375 		break;
376 #endif
377 
378 	default:
379 		/*
380 		 * Unsupported address family - fail.
381 		 */
382 		return (NULL);
383 	}
384 
385 	rpcb = clnt_dg_create(so, (struct sockaddr *)&ss,
386 	    RPCBPROG, rpcvers, 0, 0);
387 	if (!rpcb)
388 		return (NULL);
389 
390 try_tcp:
391 	parms.r_prog = prog;
392 	parms.r_vers = vers;
393 	if (do_tcp)
394 		parms.r_netid = "tcp";
395 	else
396 		parms.r_netid = "udp";
397 	parms.r_addr = "";
398 	parms.r_owner = "";
399 
400 	/*
401 	 * Use the default timeout.
402 	 */
403 	timo.tv_sec = 25;
404 	timo.tv_usec = 0;
405 again:
406 	switch (rpcvers) {
407 	case RPCBVERS4:
408 	case RPCBVERS:
409 		/*
410 		 * Try RPCBIND 4 then 3.
411 		 */
412 		uaddr = NULL;
413 		stat = CLNT_CALL(rpcb, (rpcprog_t) RPCBPROC_GETADDR,
414 		    (xdrproc_t) xdr_rpcb, &parms,
415 		    (xdrproc_t) xdr_wrapstring, &uaddr, timo);
416 		if (stat == RPC_SUCCESS) {
417 			/*
418 			 * We have a reply from the remote RPCBIND - turn it
419 			 * into an appropriate address and make a new client
420 			 * that can talk to the remote NLM.
421 			 *
422 			 * XXX fixup IPv6 scope ID.
423 			 */
424 			struct netbuf *a;
425 			a = __rpc_uaddr2taddr_af(ss.ss_family, uaddr);
426 			if (!a) {
427 				tryagain = TRUE;
428 			} else {
429 				tryagain = FALSE;
430 				memcpy(&ss, a->buf, a->len);
431 				free(a->buf, M_RPC);
432 				free(a, M_RPC);
433 				xdr_free((xdrproc_t) xdr_wrapstring, &uaddr);
434 			}
435 		}
436 		if (tryagain || stat == RPC_PROGVERSMISMATCH) {
437 			if (rpcvers == RPCBVERS4)
438 				rpcvers = RPCBVERS;
439 			else if (rpcvers == RPCBVERS)
440 				rpcvers = PMAPVERS;
441 			CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers);
442 			goto again;
443 		}
444 		break;
445 	case PMAPVERS:
446 		/*
447 		 * Try portmap.
448 		 */
449 		mapping.pm_prog = parms.r_prog;
450 		mapping.pm_vers = parms.r_vers;
451 		mapping.pm_prot = do_tcp ? IPPROTO_TCP : IPPROTO_UDP;
452 		mapping.pm_port = 0;
453 
454 		stat = CLNT_CALL(rpcb, (rpcprog_t) PMAPPROC_GETPORT,
455 		    (xdrproc_t) xdr_portmap, &mapping,
456 		    (xdrproc_t) xdr_u_short, &port, timo);
457 
458 		if (stat == RPC_SUCCESS) {
459 			switch (ss.ss_family) {
460 			case AF_INET:
461 				((struct sockaddr_in *)&ss)->sin_port =
462 					htons(port);
463 				break;
464 
465 #ifdef INET6
466 			case AF_INET6:
467 				((struct sockaddr_in6 *)&ss)->sin6_port =
468 					htons(port);
469 				break;
470 #endif
471 			}
472 		}
473 		break;
474 	default:
475 		panic("invalid rpcvers %d", rpcvers);
476 	}
477 	/*
478 	 * We may have a positive response from the portmapper, but the NLM
479 	 * service was not found. Make sure we received a valid port.
480 	 */
481 	switch (ss.ss_family) {
482 	case AF_INET:
483 		port = ((struct sockaddr_in *)&ss)->sin_port;
484 		break;
485 #ifdef INET6
486 	case AF_INET6:
487 		port = ((struct sockaddr_in6 *)&ss)->sin6_port;
488 		break;
489 #endif
490 	}
491 	if (stat != RPC_SUCCESS || !port) {
492 		/*
493 		 * If we were able to talk to rpcbind or portmap, but the udp
494 		 * variant wasn't available, ask about tcp.
495 		 *
496 		 * XXX - We could also check for a TCP portmapper, but
497 		 * if the host is running a portmapper at all, we should be able
498 		 * to hail it over UDP.
499 		 */
500 		if (stat == RPC_SUCCESS && !do_tcp) {
501 			do_tcp = TRUE;
502 			goto try_tcp;
503 		}
504 
505 		/* Otherwise, bad news. */
506 		NLM_ERR("NLM: failed to contact remote rpcbind, "
507 		    "stat = %d, port = %d\n", (int) stat, port);
508 		CLNT_DESTROY(rpcb);
509 		return (NULL);
510 	}
511 
512 	if (do_tcp) {
513 		/*
514 		 * Destroy the UDP client we used to speak to rpcbind and
515 		 * recreate as a TCP client.
516 		 */
517 		struct netconfig *nconf = NULL;
518 
519 		CLNT_DESTROY(rpcb);
520 
521 		switch (ss.ss_family) {
522 		case AF_INET:
523 			nconf = getnetconfigent("tcp");
524 			break;
525 #ifdef INET6
526 		case AF_INET6:
527 			nconf = getnetconfigent("tcp6");
528 			break;
529 #endif
530 		}
531 
532 		rpcb = clnt_reconnect_create(nconf, (struct sockaddr *)&ss,
533 		    prog, vers, 0, 0);
534 		CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan);
535 		rpcb->cl_auth = nlm_auth;
536 
537 	} else {
538 		/*
539 		 * Re-use the client we used to speak to rpcbind.
540 		 */
541 		CLNT_CONTROL(rpcb, CLSET_SVC_ADDR, &ss);
542 		CLNT_CONTROL(rpcb, CLSET_PROG, &prog);
543 		CLNT_CONTROL(rpcb, CLSET_VERS, &vers);
544 		CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan);
545 		rpcb->cl_auth = nlm_auth;
546 	}
547 
548 	return (rpcb);
549 }
550 
551 /*
552  * This async callback after when an async lock request has been
553  * granted. We notify the host which initiated the request.
554  */
555 static void
556 nlm_lock_callback(void *arg, int pending)
557 {
558 	struct nlm_async_lock *af = (struct nlm_async_lock *) arg;
559 	struct rpc_callextra ext;
560 
561 	NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) granted,"
562 	    " cookie %d:%d\n", af, af->af_host->nh_caller_name,
563 	    af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie),
564 	    ng_cookie(&af->af_granted.cookie));
565 
566 	/*
567 	 * Send the results back to the host.
568 	 *
569 	 * Note: there is a possible race here with nlm_host_notify
570 	 * destroying the RPC client. To avoid problems, the first
571 	 * thing nlm_host_notify does is to cancel pending async lock
572 	 * requests.
573 	 */
574 	memset(&ext, 0, sizeof(ext));
575 	ext.rc_auth = nlm_auth;
576 	if (af->af_host->nh_vers == NLM_VERS4) {
577 		nlm4_granted_msg_4(&af->af_granted,
578 		    NULL, af->af_rpc, &ext, nlm_zero_tv);
579 	} else {
580 		/*
581 		 * Back-convert to legacy protocol
582 		 */
583 		nlm_testargs granted;
584 		granted.cookie = af->af_granted.cookie;
585 		granted.exclusive = af->af_granted.exclusive;
586 		granted.alock.caller_name =
587 			af->af_granted.alock.caller_name;
588 		granted.alock.fh = af->af_granted.alock.fh;
589 		granted.alock.oh = af->af_granted.alock.oh;
590 		granted.alock.svid = af->af_granted.alock.svid;
591 		granted.alock.l_offset =
592 			af->af_granted.alock.l_offset;
593 		granted.alock.l_len =
594 			af->af_granted.alock.l_len;
595 
596 		nlm_granted_msg_1(&granted,
597 		    NULL, af->af_rpc, &ext, nlm_zero_tv);
598 	}
599 
600 	/*
601 	 * Move this entry to the nh_granted list.
602 	 */
603 	af->af_expiretime = time_uptime + NLM_EXPIRE_TIMEOUT;
604 	mtx_lock(&af->af_host->nh_lock);
605 	TAILQ_REMOVE(&af->af_host->nh_pending, af, af_link);
606 	TAILQ_INSERT_TAIL(&af->af_host->nh_granted, af, af_link);
607 	mtx_unlock(&af->af_host->nh_lock);
608 }
609 
610 /*
611  * Free an async lock request. The request must have been removed from
612  * any list.
613  */
614 static void
615 nlm_free_async_lock(struct nlm_async_lock *af)
616 {
617 	/*
618 	 * Free an async lock.
619 	 */
620 	if (af->af_rpc)
621 		CLNT_RELEASE(af->af_rpc);
622 	xdr_free((xdrproc_t) xdr_nlm4_testargs, &af->af_granted);
623 	if (af->af_vp)
624 		vrele(af->af_vp);
625 	free(af, M_NLM);
626 }
627 
628 /*
629  * Cancel our async request - this must be called with
630  * af->nh_host->nh_lock held. This is slightly complicated by a
631  * potential race with our own callback. If we fail to cancel the
632  * lock, it must already have been granted - we make sure our async
633  * task has completed by calling taskqueue_drain in this case.
634  */
635 static int
636 nlm_cancel_async_lock(struct nlm_async_lock *af)
637 {
638 	struct nlm_host *host = af->af_host;
639 	int error;
640 
641 	mtx_assert(&host->nh_lock, MA_OWNED);
642 
643 	mtx_unlock(&host->nh_lock);
644 
645 	error = VOP_ADVLOCKASYNC(af->af_vp, NULL, F_CANCEL, &af->af_fl,
646 	    F_REMOTE, NULL, &af->af_cookie);
647 
648 	if (error) {
649 		/*
650 		 * We failed to cancel - make sure our callback has
651 		 * completed before we continue.
652 		 */
653 		taskqueue_drain(taskqueue_thread, &af->af_task);
654 	}
655 
656 	mtx_lock(&host->nh_lock);
657 
658 	if (!error) {
659 		NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) "
660 		    "cancelled\n", af, host->nh_caller_name, host->nh_sysid);
661 
662 		/*
663 		 * Remove from the nh_pending list and free now that
664 		 * we are safe from the callback.
665 		 */
666 		TAILQ_REMOVE(&host->nh_pending, af, af_link);
667 		mtx_unlock(&host->nh_lock);
668 		nlm_free_async_lock(af);
669 		mtx_lock(&host->nh_lock);
670 	}
671 
672 	return (error);
673 }
674 
675 static void
676 nlm_check_expired_locks(struct nlm_host *host)
677 {
678 	struct nlm_async_lock *af;
679 	time_t uptime = time_uptime;
680 
681 	mtx_lock(&host->nh_lock);
682 	while ((af = TAILQ_FIRST(&host->nh_granted)) != NULL
683 	    && uptime >= af->af_expiretime) {
684 		NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) expired,"
685 		    " cookie %d:%d\n", af, af->af_host->nh_caller_name,
686 		    af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie),
687 		    ng_cookie(&af->af_granted.cookie));
688 		TAILQ_REMOVE(&host->nh_granted, af, af_link);
689 		mtx_unlock(&host->nh_lock);
690 		nlm_free_async_lock(af);
691 		mtx_lock(&host->nh_lock);
692 	}
693 	while ((af = TAILQ_FIRST(&host->nh_finished)) != NULL) {
694 		TAILQ_REMOVE(&host->nh_finished, af, af_link);
695 		mtx_unlock(&host->nh_lock);
696 		nlm_free_async_lock(af);
697 		mtx_lock(&host->nh_lock);
698 	}
699 	mtx_unlock(&host->nh_lock);
700 }
701 
702 /*
703  * Free resources used by a host. This is called after the reference
704  * count has reached zero so it doesn't need to worry about locks.
705  */
706 static void
707 nlm_host_destroy(struct nlm_host *host)
708 {
709 
710 	mtx_lock(&nlm_global_lock);
711 	TAILQ_REMOVE(&nlm_hosts, host, nh_link);
712 	mtx_unlock(&nlm_global_lock);
713 
714 	if (host->nh_srvrpc.nr_client)
715 		CLNT_RELEASE(host->nh_srvrpc.nr_client);
716 	if (host->nh_clntrpc.nr_client)
717 		CLNT_RELEASE(host->nh_clntrpc.nr_client);
718 	mtx_destroy(&host->nh_lock);
719 	sysctl_ctx_free(&host->nh_sysctl);
720 	free(host, M_NLM);
721 }
722 
723 /*
724  * Thread start callback for client lock recovery
725  */
726 static void
727 nlm_client_recovery_start(void *arg)
728 {
729 	struct nlm_host *host = (struct nlm_host *) arg;
730 
731 	NLM_DEBUG(1, "NLM: client lock recovery for %s started\n",
732 	    host->nh_caller_name);
733 
734 	nlm_client_recovery(host);
735 
736 	NLM_DEBUG(1, "NLM: client lock recovery for %s completed\n",
737 	    host->nh_caller_name);
738 
739 	host->nh_monstate = NLM_MONITORED;
740 	nlm_host_release(host);
741 
742 	kthread_exit();
743 }
744 
745 /*
746  * This is called when we receive a host state change notification. We
747  * unlock any active locks owned by the host. When rpc.lockd is
748  * shutting down, this function is called with newstate set to zero
749  * which allows us to cancel any pending async locks and clear the
750  * locking state.
751  */
752 static void
753 nlm_host_notify(struct nlm_host *host, int newstate)
754 {
755 	struct nlm_async_lock *af;
756 
757 	if (newstate) {
758 		NLM_DEBUG(1, "NLM: host %s (sysid %d) rebooted, new "
759 		    "state is %d\n", host->nh_caller_name,
760 		    host->nh_sysid, newstate);
761 	}
762 
763 	/*
764 	 * Cancel any pending async locks for this host.
765 	 */
766 	mtx_lock(&host->nh_lock);
767 	while ((af = TAILQ_FIRST(&host->nh_pending)) != NULL) {
768 		/*
769 		 * nlm_cancel_async_lock will remove the entry from
770 		 * nh_pending and free it.
771 		 */
772 		nlm_cancel_async_lock(af);
773 	}
774 	mtx_unlock(&host->nh_lock);
775 	nlm_check_expired_locks(host);
776 
777 	/*
778 	 * The host just rebooted - trash its locks.
779 	 */
780 	lf_clearremotesys(host->nh_sysid);
781 	host->nh_state = newstate;
782 
783 	/*
784 	 * If we have any remote locks for this host (i.e. it
785 	 * represents a remote NFS server that our local NFS client
786 	 * has locks for), start a recovery thread.
787 	 */
788 	if (newstate != 0
789 	    && host->nh_monstate != NLM_RECOVERING
790 	    && lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid) > 0) {
791 		struct thread *td;
792 		host->nh_monstate = NLM_RECOVERING;
793 		refcount_acquire(&host->nh_refs);
794 		kthread_add(nlm_client_recovery_start, host, curproc, &td, 0, 0,
795 		    "NFS lock recovery for %s", host->nh_caller_name);
796 	}
797 }
798 
799 /*
800  * Sysctl handler to count the number of locks for a sysid.
801  */
802 static int
803 nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
804 {
805 	struct nlm_host *host;
806 	int count;
807 
808 	host = oidp->oid_arg1;
809 	count = lf_countlocks(host->nh_sysid);
810 	return sysctl_handle_int(oidp, &count, 0, req);
811 }
812 
813 /*
814  * Sysctl handler to count the number of client locks for a sysid.
815  */
816 static int
817 nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
818 {
819 	struct nlm_host *host;
820 	int count;
821 
822 	host = oidp->oid_arg1;
823 	count = lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid);
824 	return sysctl_handle_int(oidp, &count, 0, req);
825 }
826 
827 /*
828  * Create a new NLM host.
829  */
830 static struct nlm_host *
831 nlm_create_host(const char* caller_name)
832 {
833 	struct nlm_host *host;
834 	struct sysctl_oid *oid;
835 
836 	mtx_assert(&nlm_global_lock, MA_OWNED);
837 
838 	NLM_DEBUG(1, "NLM: new host %s (sysid %d)\n",
839 	    caller_name, nlm_next_sysid);
840 	host = malloc(sizeof(struct nlm_host), M_NLM, M_NOWAIT|M_ZERO);
841 	if (!host)
842 		return (NULL);
843 	mtx_init(&host->nh_lock, "nh_lock", NULL, MTX_DEF);
844 	host->nh_refs = 1;
845 	strlcpy(host->nh_caller_name, caller_name, MAXNAMELEN);
846 	host->nh_sysid = nlm_next_sysid++;
847 	snprintf(host->nh_sysid_string, sizeof(host->nh_sysid_string),
848 		"%d", host->nh_sysid);
849 	host->nh_vers = 0;
850 	host->nh_state = 0;
851 	host->nh_monstate = NLM_UNMONITORED;
852 	host->nh_grantcookie = 1;
853 	TAILQ_INIT(&host->nh_pending);
854 	TAILQ_INIT(&host->nh_granted);
855 	TAILQ_INIT(&host->nh_finished);
856 	TAILQ_INSERT_TAIL(&nlm_hosts, host, nh_link);
857 
858 	mtx_unlock(&nlm_global_lock);
859 
860 	sysctl_ctx_init(&host->nh_sysctl);
861 	oid = SYSCTL_ADD_NODE(&host->nh_sysctl,
862 	    SYSCTL_STATIC_CHILDREN(_vfs_nlm_sysid),
863 	    OID_AUTO, host->nh_sysid_string, CTLFLAG_RD, NULL, "");
864 	SYSCTL_ADD_STRING(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
865 	    "hostname", CTLFLAG_RD, host->nh_caller_name, 0, "");
866 	SYSCTL_ADD_UINT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
867 	    "version", CTLFLAG_RD, &host->nh_vers, 0, "");
868 	SYSCTL_ADD_UINT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
869 	    "monitored", CTLFLAG_RD, &host->nh_monstate, 0, "");
870 	SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
871 	    "lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0,
872 	    nlm_host_lock_count_sysctl, "I", "");
873 	SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
874 	    "client_lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0,
875 	    nlm_host_client_lock_count_sysctl, "I", "");
876 
877 	mtx_lock(&nlm_global_lock);
878 
879 	return (host);
880 }
881 
882 /*
883  * Acquire the next sysid for remote locks not handled by the NLM.
884  */
885 uint32_t
886 nlm_acquire_next_sysid(void)
887 {
888 	uint32_t next_sysid;
889 
890 	mtx_lock(&nlm_global_lock);
891 	next_sysid = nlm_next_sysid++;
892 	mtx_unlock(&nlm_global_lock);
893 	return (next_sysid);
894 }
895 
896 /*
897  * Return non-zero if the address parts of the two sockaddrs are the
898  * same.
899  */
900 static int
901 nlm_compare_addr(const struct sockaddr *a, const struct sockaddr *b)
902 {
903 	const struct sockaddr_in *a4, *b4;
904 #ifdef INET6
905 	const struct sockaddr_in6 *a6, *b6;
906 #endif
907 
908 	if (a->sa_family != b->sa_family)
909 		return (FALSE);
910 
911 	switch (a->sa_family) {
912 	case AF_INET:
913 		a4 = (const struct sockaddr_in *) a;
914 		b4 = (const struct sockaddr_in *) b;
915 		return !memcmp(&a4->sin_addr, &b4->sin_addr,
916 		    sizeof(a4->sin_addr));
917 #ifdef INET6
918 	case AF_INET6:
919 		a6 = (const struct sockaddr_in6 *) a;
920 		b6 = (const struct sockaddr_in6 *) b;
921 		return !memcmp(&a6->sin6_addr, &b6->sin6_addr,
922 		    sizeof(a6->sin6_addr));
923 #endif
924 	}
925 
926 	return (0);
927 }
928 
929 /*
930  * Check for idle hosts and stop monitoring them. We could also free
931  * the host structure here, possibly after a larger timeout but that
932  * would require some care to avoid races with
933  * e.g. nlm_host_lock_count_sysctl.
934  */
935 static void
936 nlm_check_idle(void)
937 {
938 	struct nlm_host *host;
939 
940 	mtx_assert(&nlm_global_lock, MA_OWNED);
941 
942 	if (time_uptime <= nlm_next_idle_check)
943 		return;
944 
945 	nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
946 
947 	TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
948 		if (host->nh_monstate == NLM_MONITORED
949 		    && time_uptime > host->nh_idle_timeout) {
950 			mtx_unlock(&nlm_global_lock);
951 			if (lf_countlocks(host->nh_sysid) > 0
952 			    || lf_countlocks(NLM_SYSID_CLIENT
953 				+ host->nh_sysid)) {
954 				host->nh_idle_timeout =
955 					time_uptime + NLM_IDLE_TIMEOUT;
956 				mtx_lock(&nlm_global_lock);
957 				continue;
958 			}
959 			nlm_host_unmonitor(host);
960 			mtx_lock(&nlm_global_lock);
961 		}
962 	}
963 }
964 
965 /*
966  * Search for an existing NLM host that matches the given name
967  * (typically the caller_name element of an nlm4_lock).  If none is
968  * found, create a new host. If 'addr' is non-NULL, record the remote
969  * address of the host so that we can call it back for async
970  * responses. If 'vers' is greater than zero then record the NLM
971  * program version to use to communicate with this client.
972  */
973 struct nlm_host *
974 nlm_find_host_by_name(const char *name, const struct sockaddr *addr,
975     rpcvers_t vers)
976 {
977 	struct nlm_host *host;
978 
979 	mtx_lock(&nlm_global_lock);
980 
981 	/*
982 	 * The remote host is determined by caller_name.
983 	 */
984 	TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
985 		if (!strcmp(host->nh_caller_name, name))
986 			break;
987 	}
988 
989 	if (!host) {
990 		host = nlm_create_host(name);
991 		if (!host) {
992 			mtx_unlock(&nlm_global_lock);
993 			return (NULL);
994 		}
995 	}
996 	refcount_acquire(&host->nh_refs);
997 
998 	host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
999 
1000 	/*
1001 	 * If we have an address for the host, record it so that we
1002 	 * can send async replies etc.
1003 	 */
1004 	if (addr) {
1005 
1006 		KASSERT(addr->sa_len < sizeof(struct sockaddr_storage),
1007 		    ("Strange remote transport address length"));
1008 
1009 		/*
1010 		 * If we have seen an address before and we currently
1011 		 * have an RPC client handle, make sure the address is
1012 		 * the same, otherwise discard the client handle.
1013 		 */
1014 		if (host->nh_addr.ss_len && host->nh_srvrpc.nr_client) {
1015 			if (!nlm_compare_addr(
1016 				    (struct sockaddr *) &host->nh_addr,
1017 				    addr)
1018 			    || host->nh_vers != vers) {
1019 				CLIENT *client;
1020 				mtx_lock(&host->nh_lock);
1021 				client = host->nh_srvrpc.nr_client;
1022 				host->nh_srvrpc.nr_client = NULL;
1023 				mtx_unlock(&host->nh_lock);
1024 				if (client) {
1025 					CLNT_RELEASE(client);
1026 				}
1027 			}
1028 		}
1029 		memcpy(&host->nh_addr, addr, addr->sa_len);
1030 		host->nh_vers = vers;
1031 	}
1032 
1033 	nlm_check_idle();
1034 
1035 	mtx_unlock(&nlm_global_lock);
1036 
1037 	return (host);
1038 }
1039 
1040 /*
1041  * Search for an existing NLM host that matches the given remote
1042  * address. If none is found, create a new host with the requested
1043  * address and remember 'vers' as the NLM protocol version to use for
1044  * that host.
1045  */
1046 struct nlm_host *
1047 nlm_find_host_by_addr(const struct sockaddr *addr, int vers)
1048 {
1049 	/*
1050 	 * Fake up a name using inet_ntop. This buffer is
1051 	 * large enough for an IPv6 address.
1052 	 */
1053 	char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"];
1054 	struct nlm_host *host;
1055 
1056 	switch (addr->sa_family) {
1057 	case AF_INET:
1058 		inet_ntop(AF_INET,
1059 		    &((const struct sockaddr_in *) addr)->sin_addr,
1060 		    tmp, sizeof tmp);
1061 		break;
1062 #ifdef INET6
1063 	case AF_INET6:
1064 		inet_ntop(AF_INET6,
1065 		    &((const struct sockaddr_in6 *) addr)->sin6_addr,
1066 		    tmp, sizeof tmp);
1067 		break;
1068 #endif
1069 	default:
1070 		strcmp(tmp, "<unknown>");
1071 	}
1072 
1073 
1074 	mtx_lock(&nlm_global_lock);
1075 
1076 	/*
1077 	 * The remote host is determined by caller_name.
1078 	 */
1079 	TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
1080 		if (nlm_compare_addr(addr,
1081 			(const struct sockaddr *) &host->nh_addr))
1082 			break;
1083 	}
1084 
1085 	if (!host) {
1086 		host = nlm_create_host(tmp);
1087 		if (!host) {
1088 			mtx_unlock(&nlm_global_lock);
1089 			return (NULL);
1090 		}
1091 		memcpy(&host->nh_addr, addr, addr->sa_len);
1092 		host->nh_vers = vers;
1093 	}
1094 	refcount_acquire(&host->nh_refs);
1095 
1096 	host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
1097 
1098 	nlm_check_idle();
1099 
1100 	mtx_unlock(&nlm_global_lock);
1101 
1102 	return (host);
1103 }
1104 
1105 /*
1106  * Find the NLM host that matches the value of 'sysid'. If none
1107  * exists, return NULL.
1108  */
1109 static struct nlm_host *
1110 nlm_find_host_by_sysid(int sysid)
1111 {
1112 	struct nlm_host *host;
1113 
1114 	TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
1115 		if (host->nh_sysid == sysid) {
1116 			refcount_acquire(&host->nh_refs);
1117 			return (host);
1118 		}
1119 	}
1120 
1121 	return (NULL);
1122 }
1123 
1124 void nlm_host_release(struct nlm_host *host)
1125 {
1126 	if (refcount_release(&host->nh_refs)) {
1127 		/*
1128 		 * Free the host
1129 		 */
1130 		nlm_host_destroy(host);
1131 	}
1132 }
1133 
1134 /*
1135  * Unregister this NLM host with the local NSM due to idleness.
1136  */
1137 static void
1138 nlm_host_unmonitor(struct nlm_host *host)
1139 {
1140 	mon_id smmonid;
1141 	sm_stat_res smstat;
1142 	struct timeval timo;
1143 	enum clnt_stat stat;
1144 
1145 	NLM_DEBUG(1, "NLM: unmonitoring %s (sysid %d)\n",
1146 	    host->nh_caller_name, host->nh_sysid);
1147 
1148 	/*
1149 	 * We put our assigned system ID value in the priv field to
1150 	 * make it simpler to find the host if we are notified of a
1151 	 * host restart.
1152 	 */
1153 	smmonid.mon_name = host->nh_caller_name;
1154 	smmonid.my_id.my_name = "localhost";
1155 	smmonid.my_id.my_prog = NLM_PROG;
1156 	smmonid.my_id.my_vers = NLM_SM;
1157 	smmonid.my_id.my_proc = NLM_SM_NOTIFY;
1158 
1159 	timo.tv_sec = 25;
1160 	timo.tv_usec = 0;
1161 	stat = CLNT_CALL(nlm_nsm, SM_UNMON,
1162 	    (xdrproc_t) xdr_mon, &smmonid,
1163 	    (xdrproc_t) xdr_sm_stat, &smstat, timo);
1164 
1165 	if (stat != RPC_SUCCESS) {
1166 		NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat);
1167 		return;
1168 	}
1169 	if (smstat.res_stat == stat_fail) {
1170 		NLM_ERR("Local NSM refuses to unmonitor %s\n",
1171 		    host->nh_caller_name);
1172 		return;
1173 	}
1174 
1175 	host->nh_monstate = NLM_UNMONITORED;
1176 }
1177 
1178 /*
1179  * Register this NLM host with the local NSM so that we can be
1180  * notified if it reboots.
1181  */
1182 void
1183 nlm_host_monitor(struct nlm_host *host, int state)
1184 {
1185 	mon smmon;
1186 	sm_stat_res smstat;
1187 	struct timeval timo;
1188 	enum clnt_stat stat;
1189 
1190 	if (state && !host->nh_state) {
1191 		/*
1192 		 * This is the first time we have seen an NSM state
1193 		 * value for this host. We record it here to help
1194 		 * detect host reboots.
1195 		 */
1196 		host->nh_state = state;
1197 		NLM_DEBUG(1, "NLM: host %s (sysid %d) has NSM state %d\n",
1198 		    host->nh_caller_name, host->nh_sysid, state);
1199 	}
1200 
1201 	mtx_lock(&host->nh_lock);
1202 	if (host->nh_monstate != NLM_UNMONITORED) {
1203 		mtx_unlock(&host->nh_lock);
1204 		return;
1205 	}
1206 	host->nh_monstate = NLM_MONITORED;
1207 	mtx_unlock(&host->nh_lock);
1208 
1209 	NLM_DEBUG(1, "NLM: monitoring %s (sysid %d)\n",
1210 	    host->nh_caller_name, host->nh_sysid);
1211 
1212 	/*
1213 	 * We put our assigned system ID value in the priv field to
1214 	 * make it simpler to find the host if we are notified of a
1215 	 * host restart.
1216 	 */
1217 	smmon.mon_id.mon_name = host->nh_caller_name;
1218 	smmon.mon_id.my_id.my_name = "localhost";
1219 	smmon.mon_id.my_id.my_prog = NLM_PROG;
1220 	smmon.mon_id.my_id.my_vers = NLM_SM;
1221 	smmon.mon_id.my_id.my_proc = NLM_SM_NOTIFY;
1222 	memcpy(smmon.priv, &host->nh_sysid, sizeof(host->nh_sysid));
1223 
1224 	timo.tv_sec = 25;
1225 	timo.tv_usec = 0;
1226 	stat = CLNT_CALL(nlm_nsm, SM_MON,
1227 	    (xdrproc_t) xdr_mon, &smmon,
1228 	    (xdrproc_t) xdr_sm_stat, &smstat, timo);
1229 
1230 	if (stat != RPC_SUCCESS) {
1231 		NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat);
1232 		return;
1233 	}
1234 	if (smstat.res_stat == stat_fail) {
1235 		NLM_ERR("Local NSM refuses to monitor %s\n",
1236 		    host->nh_caller_name);
1237 		mtx_lock(&host->nh_lock);
1238 		host->nh_monstate = NLM_MONITOR_FAILED;
1239 		mtx_unlock(&host->nh_lock);
1240 		return;
1241 	}
1242 
1243 	host->nh_monstate = NLM_MONITORED;
1244 }
1245 
1246 /*
1247  * Return an RPC client handle that can be used to talk to the NLM
1248  * running on the given host.
1249  */
1250 CLIENT *
1251 nlm_host_get_rpc(struct nlm_host *host, bool_t isserver)
1252 {
1253 	struct nlm_rpc *rpc;
1254 	CLIENT *client;
1255 
1256 	mtx_lock(&host->nh_lock);
1257 
1258 	if (isserver)
1259 		rpc = &host->nh_srvrpc;
1260 	else
1261 		rpc = &host->nh_clntrpc;
1262 
1263 	/*
1264 	 * We can't hold onto RPC handles for too long - the async
1265 	 * call/reply protocol used by some NLM clients makes it hard
1266 	 * to tell when they change port numbers (e.g. after a
1267 	 * reboot). Note that if a client reboots while it isn't
1268 	 * holding any locks, it won't bother to notify us. We
1269 	 * expire the RPC handles after two minutes.
1270 	 */
1271 	if (rpc->nr_client && time_uptime > rpc->nr_create_time + 2*60) {
1272 		client = rpc->nr_client;
1273 		rpc->nr_client = NULL;
1274 		mtx_unlock(&host->nh_lock);
1275 		CLNT_RELEASE(client);
1276 		mtx_lock(&host->nh_lock);
1277 	}
1278 
1279 	if (!rpc->nr_client) {
1280 		mtx_unlock(&host->nh_lock);
1281 		client = nlm_get_rpc((struct sockaddr *)&host->nh_addr,
1282 		    NLM_PROG, host->nh_vers);
1283 		mtx_lock(&host->nh_lock);
1284 
1285 		if (client) {
1286 			if (rpc->nr_client) {
1287 				mtx_unlock(&host->nh_lock);
1288 				CLNT_DESTROY(client);
1289 				mtx_lock(&host->nh_lock);
1290 			} else {
1291 				rpc->nr_client = client;
1292 				rpc->nr_create_time = time_uptime;
1293 			}
1294 		}
1295 	}
1296 
1297 	client = rpc->nr_client;
1298 	if (client)
1299 		CLNT_ACQUIRE(client);
1300 	mtx_unlock(&host->nh_lock);
1301 
1302 	return (client);
1303 
1304 }
1305 
1306 int nlm_host_get_sysid(struct nlm_host *host)
1307 {
1308 
1309 	return (host->nh_sysid);
1310 }
1311 
1312 int
1313 nlm_host_get_state(struct nlm_host *host)
1314 {
1315 
1316 	return (host->nh_state);
1317 }
1318 
1319 void *
1320 nlm_register_wait_lock(struct nlm4_lock *lock, struct vnode *vp)
1321 {
1322 	struct nlm_waiting_lock *nw;
1323 
1324 	nw = malloc(sizeof(struct nlm_waiting_lock), M_NLM, M_WAITOK);
1325 	nw->nw_lock = *lock;
1326 	memcpy(&nw->nw_fh.fh_bytes, nw->nw_lock.fh.n_bytes,
1327 	    nw->nw_lock.fh.n_len);
1328 	nw->nw_lock.fh.n_bytes = nw->nw_fh.fh_bytes;
1329 	nw->nw_waiting = TRUE;
1330 	nw->nw_vp = vp;
1331 	mtx_lock(&nlm_global_lock);
1332 	TAILQ_INSERT_TAIL(&nlm_waiting_locks, nw, nw_link);
1333 	mtx_unlock(&nlm_global_lock);
1334 
1335 	return nw;
1336 }
1337 
1338 void
1339 nlm_deregister_wait_lock(void *handle)
1340 {
1341 	struct nlm_waiting_lock *nw = handle;
1342 
1343 	mtx_lock(&nlm_global_lock);
1344 	TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link);
1345 	mtx_unlock(&nlm_global_lock);
1346 
1347 	free(nw, M_NLM);
1348 }
1349 
1350 int
1351 nlm_wait_lock(void *handle, int timo)
1352 {
1353 	struct nlm_waiting_lock *nw = handle;
1354 	int error;
1355 
1356 	/*
1357 	 * If the granted message arrived before we got here,
1358 	 * nw->nw_waiting will be FALSE - in that case, don't sleep.
1359 	 */
1360 	mtx_lock(&nlm_global_lock);
1361 	error = 0;
1362 	if (nw->nw_waiting)
1363 		error = msleep(nw, &nlm_global_lock, PCATCH, "nlmlock", timo);
1364 	TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link);
1365 	if (error) {
1366 		/*
1367 		 * The granted message may arrive after the
1368 		 * interrupt/timeout but before we manage to lock the
1369 		 * mutex. Detect this by examining nw_lock.
1370 		 */
1371 		if (!nw->nw_waiting)
1372 			error = 0;
1373 	} else {
1374 		/*
1375 		 * If nlm_cancel_wait is called, then error will be
1376 		 * zero but nw_waiting will still be TRUE. We
1377 		 * translate this into EINTR.
1378 		 */
1379 		if (nw->nw_waiting)
1380 			error = EINTR;
1381 	}
1382 	mtx_unlock(&nlm_global_lock);
1383 
1384 	free(nw, M_NLM);
1385 
1386 	return (error);
1387 }
1388 
1389 void
1390 nlm_cancel_wait(struct vnode *vp)
1391 {
1392 	struct nlm_waiting_lock *nw;
1393 
1394 	mtx_lock(&nlm_global_lock);
1395 	TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
1396 		if (nw->nw_vp == vp) {
1397 			wakeup(nw);
1398 		}
1399 	}
1400 	mtx_unlock(&nlm_global_lock);
1401 }
1402 
1403 
1404 /**********************************************************************/
1405 
1406 /*
1407  * Syscall interface with userland.
1408  */
1409 
1410 extern void nlm_prog_0(struct svc_req *rqstp, SVCXPRT *transp);
1411 extern void nlm_prog_1(struct svc_req *rqstp, SVCXPRT *transp);
1412 extern void nlm_prog_3(struct svc_req *rqstp, SVCXPRT *transp);
1413 extern void nlm_prog_4(struct svc_req *rqstp, SVCXPRT *transp);
1414 
1415 static int
1416 nlm_register_services(SVCPOOL *pool, int addr_count, char **addrs)
1417 {
1418 	static rpcvers_t versions[] = {
1419 		NLM_SM, NLM_VERS, NLM_VERSX, NLM_VERS4
1420 	};
1421 	static void (*dispatchers[])(struct svc_req *, SVCXPRT *) = {
1422 		nlm_prog_0, nlm_prog_1, nlm_prog_3, nlm_prog_4
1423 	};
1424 	static const int version_count = sizeof(versions) / sizeof(versions[0]);
1425 
1426 	SVCXPRT **xprts;
1427 	char netid[16];
1428 	char uaddr[128];
1429 	struct netconfig *nconf;
1430 	int i, j, error;
1431 
1432 	if (!addr_count) {
1433 		NLM_ERR("NLM: no service addresses given - can't start server");
1434 		return (EINVAL);
1435 	}
1436 
1437 	xprts = malloc(addr_count * sizeof(SVCXPRT *), M_NLM, M_WAITOK|M_ZERO);
1438 	for (i = 0; i < version_count; i++) {
1439 		for (j = 0; j < addr_count; j++) {
1440 			/*
1441 			 * Create transports for the first version and
1442 			 * then just register everything else to the
1443 			 * same transports.
1444 			 */
1445 			if (i == 0) {
1446 				char *up;
1447 
1448 				error = copyin(&addrs[2*j], &up,
1449 				    sizeof(char*));
1450 				if (error)
1451 					goto out;
1452 				error = copyinstr(up, netid, sizeof(netid),
1453 				    NULL);
1454 				if (error)
1455 					goto out;
1456 				error = copyin(&addrs[2*j+1], &up,
1457 				    sizeof(char*));
1458 				if (error)
1459 					goto out;
1460 				error = copyinstr(up, uaddr, sizeof(uaddr),
1461 				    NULL);
1462 				if (error)
1463 					goto out;
1464 				nconf = getnetconfigent(netid);
1465 				if (!nconf) {
1466 					NLM_ERR("Can't lookup netid %s\n",
1467 					    netid);
1468 					error = EINVAL;
1469 					goto out;
1470 				}
1471 				xprts[j] = svc_tp_create(pool, dispatchers[i],
1472 				    NLM_PROG, versions[i], uaddr, nconf);
1473 				if (!xprts[j]) {
1474 					NLM_ERR("NLM: unable to create "
1475 					    "(NLM_PROG, %d).\n", versions[i]);
1476 					error = EINVAL;
1477 					goto out;
1478 				}
1479 				freenetconfigent(nconf);
1480 			} else {
1481 				nconf = getnetconfigent(xprts[j]->xp_netid);
1482 				rpcb_unset(NLM_PROG, versions[i], nconf);
1483 				if (!svc_reg(xprts[j], NLM_PROG, versions[i],
1484 					dispatchers[i], nconf)) {
1485 					NLM_ERR("NLM: can't register "
1486 					    "(NLM_PROG, %d)\n", versions[i]);
1487 					error = EINVAL;
1488 					goto out;
1489 				}
1490 			}
1491 		}
1492 	}
1493 	error = 0;
1494 out:
1495 	for (j = 0; j < addr_count; j++) {
1496 		if (xprts[j])
1497 			SVC_RELEASE(xprts[j]);
1498 	}
1499 	free(xprts, M_NLM);
1500 	return (error);
1501 }
1502 
1503 /*
1504  * Main server entry point. Contacts the local NSM to get its current
1505  * state and send SM_UNMON_ALL. Registers the NLM services and then
1506  * services requests. Does not return until the server is interrupted
1507  * by a signal.
1508  */
1509 static int
1510 nlm_server_main(int addr_count, char **addrs)
1511 {
1512 	struct thread *td = curthread;
1513 	int error;
1514 	SVCPOOL *pool = NULL;
1515 	struct sockopt opt;
1516 	int portlow;
1517 #ifdef INET6
1518 	struct sockaddr_in6 sin6;
1519 #endif
1520 	struct sockaddr_in sin;
1521 	my_id id;
1522 	sm_stat smstat;
1523 	struct timeval timo;
1524 	enum clnt_stat stat;
1525 	struct nlm_host *host, *nhost;
1526 	struct nlm_waiting_lock *nw;
1527 	vop_advlock_t *old_nfs_advlock;
1528 	vop_reclaim_t *old_nfs_reclaim;
1529 	int v4_used;
1530 #ifdef INET6
1531 	int v6_used;
1532 #endif
1533 
1534 	if (nlm_socket) {
1535 		NLM_ERR("NLM: can't start server - "
1536 		    "it appears to be running already\n");
1537 		return (EPERM);
1538 	}
1539 
1540 	memset(&opt, 0, sizeof(opt));
1541 
1542 	nlm_socket = NULL;
1543 	error = socreate(AF_INET, &nlm_socket, SOCK_DGRAM, 0,
1544 	    td->td_ucred, td);
1545 	if (error) {
1546 		NLM_ERR("NLM: can't create IPv4 socket - error %d\n", error);
1547 		return (error);
1548 	}
1549 	opt.sopt_dir = SOPT_SET;
1550 	opt.sopt_level = IPPROTO_IP;
1551 	opt.sopt_name = IP_PORTRANGE;
1552 	portlow = IP_PORTRANGE_LOW;
1553 	opt.sopt_val = &portlow;
1554 	opt.sopt_valsize = sizeof(portlow);
1555 	sosetopt(nlm_socket, &opt);
1556 
1557 #ifdef INET6
1558 	nlm_socket6 = NULL;
1559 	error = socreate(AF_INET6, &nlm_socket6, SOCK_DGRAM, 0,
1560 	    td->td_ucred, td);
1561 	if (error) {
1562 		NLM_ERR("NLM: can't create IPv6 socket - error %d\n", error);
1563 		goto out;
1564 		return (error);
1565 	}
1566 	opt.sopt_dir = SOPT_SET;
1567 	opt.sopt_level = IPPROTO_IPV6;
1568 	opt.sopt_name = IPV6_PORTRANGE;
1569 	portlow = IPV6_PORTRANGE_LOW;
1570 	opt.sopt_val = &portlow;
1571 	opt.sopt_valsize = sizeof(portlow);
1572 	sosetopt(nlm_socket6, &opt);
1573 #endif
1574 
1575 	nlm_auth = authunix_create(curthread->td_ucred);
1576 
1577 #ifdef INET6
1578 	memset(&sin6, 0, sizeof(sin6));
1579 	sin6.sin6_len = sizeof(sin6);
1580 	sin6.sin6_family = AF_INET6;
1581 	sin6.sin6_addr = in6addr_loopback;
1582 	nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin6, SM_PROG, SM_VERS);
1583 	if (!nlm_nsm) {
1584 #endif
1585 		memset(&sin, 0, sizeof(sin));
1586 		sin.sin_len = sizeof(sin);
1587 		sin.sin_family = AF_INET;
1588 		sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1589 		nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin, SM_PROG,
1590 		    SM_VERS);
1591 #ifdef INET6
1592 	}
1593 #endif
1594 
1595 	if (!nlm_nsm) {
1596 		NLM_ERR("Can't start NLM - unable to contact NSM\n");
1597 		error = EINVAL;
1598 		goto out;
1599 	}
1600 
1601 	pool = svcpool_create("NLM", NULL);
1602 
1603 	error = nlm_register_services(pool, addr_count, addrs);
1604 	if (error)
1605 		goto out;
1606 
1607 	memset(&id, 0, sizeof(id));
1608 	id.my_name = "NFS NLM";
1609 
1610 	timo.tv_sec = 25;
1611 	timo.tv_usec = 0;
1612 	stat = CLNT_CALL(nlm_nsm, SM_UNMON_ALL,
1613 	    (xdrproc_t) xdr_my_id, &id,
1614 	    (xdrproc_t) xdr_sm_stat, &smstat, timo);
1615 
1616 	if (stat != RPC_SUCCESS) {
1617 		struct rpc_err err;
1618 
1619 		CLNT_GETERR(nlm_nsm, &err);
1620 		NLM_ERR("NLM: unexpected error contacting NSM, "
1621 		    "stat=%d, errno=%d\n", stat, err.re_errno);
1622 		error = EINVAL;
1623 		goto out;
1624 	}
1625 
1626 	NLM_DEBUG(1, "NLM: local NSM state is %d\n", smstat.state);
1627 	nlm_nsm_state = smstat.state;
1628 
1629 	old_nfs_advlock = nfs_advlock_p;
1630 	nfs_advlock_p = nlm_advlock;
1631 	old_nfs_reclaim = nfs_reclaim_p;
1632 	nfs_reclaim_p = nlm_reclaim;
1633 
1634 	svc_run(pool);
1635 	error = 0;
1636 
1637 	nfs_advlock_p = old_nfs_advlock;
1638 	nfs_reclaim_p = old_nfs_reclaim;
1639 
1640 out:
1641 	if (pool)
1642 		svcpool_destroy(pool);
1643 
1644 	/*
1645 	 * We are finished communicating with the NSM.
1646 	 */
1647 	if (nlm_nsm) {
1648 		CLNT_RELEASE(nlm_nsm);
1649 		nlm_nsm = NULL;
1650 	}
1651 
1652 	/*
1653 	 * Trash all the existing state so that if the server
1654 	 * restarts, it gets a clean slate. This is complicated by the
1655 	 * possibility that there may be other threads trying to make
1656 	 * client locking requests.
1657 	 *
1658 	 * First we fake a client reboot notification which will
1659 	 * cancel any pending async locks and purge remote lock state
1660 	 * from the local lock manager. We release the reference from
1661 	 * nlm_hosts to the host (which may remove it from the list
1662 	 * and free it). After this phase, the only entries in the
1663 	 * nlm_host list should be from other threads performing
1664 	 * client lock requests. We arrange to defer closing the
1665 	 * sockets until the last RPC client handle is released.
1666 	 */
1667 	v4_used = 0;
1668 #ifdef INET6
1669 	v6_used = 0;
1670 #endif
1671 	mtx_lock(&nlm_global_lock);
1672 	TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
1673 		wakeup(nw);
1674 	}
1675 	TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) {
1676 		mtx_unlock(&nlm_global_lock);
1677 		nlm_host_notify(host, 0);
1678 		nlm_host_release(host);
1679 		mtx_lock(&nlm_global_lock);
1680 	}
1681 	TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) {
1682 		mtx_lock(&host->nh_lock);
1683 		if (host->nh_srvrpc.nr_client
1684 		    || host->nh_clntrpc.nr_client) {
1685 			if (host->nh_addr.ss_family == AF_INET)
1686 				v4_used++;
1687 #ifdef INET6
1688 			if (host->nh_addr.ss_family == AF_INET6)
1689 				v6_used++;
1690 #endif
1691 			/*
1692 			 * Note that the rpc over udp code copes
1693 			 * correctly with the fact that a socket may
1694 			 * be used by many rpc handles.
1695 			 */
1696 			if (host->nh_srvrpc.nr_client)
1697 				CLNT_CONTROL(host->nh_srvrpc.nr_client,
1698 				    CLSET_FD_CLOSE, 0);
1699 			if (host->nh_clntrpc.nr_client)
1700 				CLNT_CONTROL(host->nh_clntrpc.nr_client,
1701 				    CLSET_FD_CLOSE, 0);
1702 		}
1703 		mtx_unlock(&host->nh_lock);
1704 	}
1705 	mtx_unlock(&nlm_global_lock);
1706 
1707 	AUTH_DESTROY(nlm_auth);
1708 
1709 	if (!v4_used)
1710 		soclose(nlm_socket);
1711 	nlm_socket = NULL;
1712 #ifdef INET6
1713 	if (!v6_used)
1714 		soclose(nlm_socket6);
1715 	nlm_socket6 = NULL;
1716 #endif
1717 
1718 	return (error);
1719 }
1720 
1721 int
1722 sys_nlm_syscall(struct thread *td, struct nlm_syscall_args *uap)
1723 {
1724 	int error;
1725 
1726 #if __FreeBSD_version >= 700000
1727 	error = priv_check(td, PRIV_NFS_LOCKD);
1728 #else
1729 	error = suser(td);
1730 #endif
1731 	if (error)
1732 		return (error);
1733 
1734 	nlm_debug_level = uap->debug_level;
1735 	nlm_grace_threshold = time_uptime + uap->grace_period;
1736 	nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
1737 
1738 	return nlm_server_main(uap->addr_count, uap->addrs);
1739 }
1740 
1741 /**********************************************************************/
1742 
1743 /*
1744  * NLM implementation details, called from the RPC stubs.
1745  */
1746 
1747 
1748 void
1749 nlm_sm_notify(struct nlm_sm_status *argp)
1750 {
1751 	uint32_t sysid;
1752 	struct nlm_host *host;
1753 
1754 	NLM_DEBUG(3, "nlm_sm_notify(): mon_name = %s\n", argp->mon_name);
1755 	memcpy(&sysid, &argp->priv, sizeof(sysid));
1756 	host = nlm_find_host_by_sysid(sysid);
1757 	if (host) {
1758 		nlm_host_notify(host, argp->state);
1759 		nlm_host_release(host);
1760 	}
1761 }
1762 
1763 static void
1764 nlm_convert_to_fhandle_t(fhandle_t *fhp, struct netobj *p)
1765 {
1766 	memcpy(fhp, p->n_bytes, sizeof(fhandle_t));
1767 }
1768 
1769 struct vfs_state {
1770 	struct mount	*vs_mp;
1771 	struct vnode	*vs_vp;
1772 	int		vs_vnlocked;
1773 };
1774 
1775 static int
1776 nlm_get_vfs_state(struct nlm_host *host, struct svc_req *rqstp,
1777     fhandle_t *fhp, struct vfs_state *vs, accmode_t accmode)
1778 {
1779 	int error, exflags;
1780 	struct ucred *cred = NULL, *credanon = NULL;
1781 
1782 	memset(vs, 0, sizeof(*vs));
1783 
1784 	vs->vs_mp = vfs_getvfs(&fhp->fh_fsid);
1785 	if (!vs->vs_mp) {
1786 		return (ESTALE);
1787 	}
1788 
1789 	/* accmode == 0 means don't check, since it is an unlock. */
1790 	if (accmode != 0) {
1791 		error = VFS_CHECKEXP(vs->vs_mp,
1792 		    (struct sockaddr *)&host->nh_addr, &exflags, &credanon,
1793 		    NULL, NULL);
1794 		if (error)
1795 			goto out;
1796 
1797 		if (exflags & MNT_EXRDONLY ||
1798 		    (vs->vs_mp->mnt_flag & MNT_RDONLY)) {
1799 			error = EROFS;
1800 			goto out;
1801 		}
1802 	}
1803 
1804 	error = VFS_FHTOVP(vs->vs_mp, &fhp->fh_fid, LK_EXCLUSIVE, &vs->vs_vp);
1805 	if (error)
1806 		goto out;
1807 	vs->vs_vnlocked = TRUE;
1808 
1809 	if (accmode != 0) {
1810 		if (!svc_getcred(rqstp, &cred, NULL)) {
1811 			error = EINVAL;
1812 			goto out;
1813 		}
1814 		if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1815 			crfree(cred);
1816 			cred = credanon;
1817 			credanon = NULL;
1818 		}
1819 
1820 		/*
1821 		 * Check cred.
1822 		 */
1823 		error = VOP_ACCESS(vs->vs_vp, accmode, cred, curthread);
1824 		/*
1825 		 * If this failed and accmode != VWRITE, try again with
1826 		 * VWRITE to maintain backwards compatibility with the
1827 		 * old code that always used VWRITE.
1828 		 */
1829 		if (error != 0 && accmode != VWRITE)
1830 			error = VOP_ACCESS(vs->vs_vp, VWRITE, cred, curthread);
1831 		if (error)
1832 			goto out;
1833 	}
1834 
1835 #if __FreeBSD_version < 800011
1836 	VOP_UNLOCK(vs->vs_vp, 0, curthread);
1837 #else
1838 	VOP_UNLOCK(vs->vs_vp, 0);
1839 #endif
1840 	vs->vs_vnlocked = FALSE;
1841 
1842 out:
1843 	if (cred)
1844 		crfree(cred);
1845 	if (credanon)
1846 		crfree(credanon);
1847 
1848 	return (error);
1849 }
1850 
1851 static void
1852 nlm_release_vfs_state(struct vfs_state *vs)
1853 {
1854 
1855 	if (vs->vs_vp) {
1856 		if (vs->vs_vnlocked)
1857 			vput(vs->vs_vp);
1858 		else
1859 			vrele(vs->vs_vp);
1860 	}
1861 	if (vs->vs_mp)
1862 		vfs_rel(vs->vs_mp);
1863 }
1864 
1865 static nlm4_stats
1866 nlm_convert_error(int error)
1867 {
1868 
1869 	if (error == ESTALE)
1870 		return nlm4_stale_fh;
1871 	else if (error == EROFS)
1872 		return nlm4_rofs;
1873 	else
1874 		return nlm4_failed;
1875 }
1876 
1877 int
1878 nlm_do_test(nlm4_testargs *argp, nlm4_testres *result, struct svc_req *rqstp,
1879 	CLIENT **rpcp)
1880 {
1881 	fhandle_t fh;
1882 	struct vfs_state vs;
1883 	struct nlm_host *host, *bhost;
1884 	int error, sysid;
1885 	struct flock fl;
1886 	accmode_t accmode;
1887 
1888 	memset(result, 0, sizeof(*result));
1889 	memset(&vs, 0, sizeof(vs));
1890 
1891 	host = nlm_find_host_by_name(argp->alock.caller_name,
1892 	    svc_getrpccaller(rqstp), rqstp->rq_vers);
1893 	if (!host) {
1894 		result->stat.stat = nlm4_denied_nolocks;
1895 		return (ENOMEM);
1896 	}
1897 
1898 	NLM_DEBUG(3, "nlm_do_test(): caller_name = %s (sysid = %d)\n",
1899 	    host->nh_caller_name, host->nh_sysid);
1900 
1901 	nlm_check_expired_locks(host);
1902 	sysid = host->nh_sysid;
1903 
1904 	nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
1905 	nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
1906 
1907 	if (time_uptime < nlm_grace_threshold) {
1908 		result->stat.stat = nlm4_denied_grace_period;
1909 		goto out;
1910 	}
1911 
1912 	accmode = argp->exclusive ? VWRITE : VREAD;
1913 	error = nlm_get_vfs_state(host, rqstp, &fh, &vs, accmode);
1914 	if (error) {
1915 		result->stat.stat = nlm_convert_error(error);
1916 		goto out;
1917 	}
1918 
1919 	fl.l_start = argp->alock.l_offset;
1920 	fl.l_len = argp->alock.l_len;
1921 	fl.l_pid = argp->alock.svid;
1922 	fl.l_sysid = sysid;
1923 	fl.l_whence = SEEK_SET;
1924 	if (argp->exclusive)
1925 		fl.l_type = F_WRLCK;
1926 	else
1927 		fl.l_type = F_RDLCK;
1928 	error = VOP_ADVLOCK(vs.vs_vp, NULL, F_GETLK, &fl, F_REMOTE);
1929 	if (error) {
1930 		result->stat.stat = nlm4_failed;
1931 		goto out;
1932 	}
1933 
1934 	if (fl.l_type == F_UNLCK) {
1935 		result->stat.stat = nlm4_granted;
1936 	} else {
1937 		result->stat.stat = nlm4_denied;
1938 		result->stat.nlm4_testrply_u.holder.exclusive =
1939 			(fl.l_type == F_WRLCK);
1940 		result->stat.nlm4_testrply_u.holder.svid = fl.l_pid;
1941 		bhost = nlm_find_host_by_sysid(fl.l_sysid);
1942 		if (bhost) {
1943 			/*
1944 			 * We don't have any useful way of recording
1945 			 * the value of oh used in the original lock
1946 			 * request. Ideally, the test reply would have
1947 			 * a space for the owning host's name allowing
1948 			 * our caller's NLM to keep track.
1949 			 *
1950 			 * As far as I can see, Solaris uses an eight
1951 			 * byte structure for oh which contains a four
1952 			 * byte pid encoded in local byte order and
1953 			 * the first four bytes of the host
1954 			 * name. Linux uses a variable length string
1955 			 * 'pid@hostname' in ascii but doesn't even
1956 			 * return that in test replies.
1957 			 *
1958 			 * For the moment, return nothing in oh
1959 			 * (already zero'ed above).
1960 			 */
1961 			nlm_host_release(bhost);
1962 		}
1963 		result->stat.nlm4_testrply_u.holder.l_offset = fl.l_start;
1964 		result->stat.nlm4_testrply_u.holder.l_len = fl.l_len;
1965 	}
1966 
1967 out:
1968 	nlm_release_vfs_state(&vs);
1969 	if (rpcp)
1970 		*rpcp = nlm_host_get_rpc(host, TRUE);
1971 	nlm_host_release(host);
1972 	return (0);
1973 }
1974 
1975 int
1976 nlm_do_lock(nlm4_lockargs *argp, nlm4_res *result, struct svc_req *rqstp,
1977     bool_t monitor, CLIENT **rpcp)
1978 {
1979 	fhandle_t fh;
1980 	struct vfs_state vs;
1981 	struct nlm_host *host;
1982 	int error, sysid;
1983 	struct flock fl;
1984 	accmode_t accmode;
1985 
1986 	memset(result, 0, sizeof(*result));
1987 	memset(&vs, 0, sizeof(vs));
1988 
1989 	host = nlm_find_host_by_name(argp->alock.caller_name,
1990 	    svc_getrpccaller(rqstp), rqstp->rq_vers);
1991 	if (!host) {
1992 		result->stat.stat = nlm4_denied_nolocks;
1993 		return (ENOMEM);
1994 	}
1995 
1996 	NLM_DEBUG(3, "nlm_do_lock(): caller_name = %s (sysid = %d)\n",
1997 	    host->nh_caller_name, host->nh_sysid);
1998 
1999 	if (monitor && host->nh_state && argp->state
2000 	    && host->nh_state != argp->state) {
2001 		/*
2002 		 * The host rebooted without telling us. Trash its
2003 		 * locks.
2004 		 */
2005 		nlm_host_notify(host, argp->state);
2006 	}
2007 
2008 	nlm_check_expired_locks(host);
2009 	sysid = host->nh_sysid;
2010 
2011 	nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
2012 	nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2013 
2014 	if (time_uptime < nlm_grace_threshold && !argp->reclaim) {
2015 		result->stat.stat = nlm4_denied_grace_period;
2016 		goto out;
2017 	}
2018 
2019 	accmode = argp->exclusive ? VWRITE : VREAD;
2020 	error = nlm_get_vfs_state(host, rqstp, &fh, &vs, accmode);
2021 	if (error) {
2022 		result->stat.stat = nlm_convert_error(error);
2023 		goto out;
2024 	}
2025 
2026 	fl.l_start = argp->alock.l_offset;
2027 	fl.l_len = argp->alock.l_len;
2028 	fl.l_pid = argp->alock.svid;
2029 	fl.l_sysid = sysid;
2030 	fl.l_whence = SEEK_SET;
2031 	if (argp->exclusive)
2032 		fl.l_type = F_WRLCK;
2033 	else
2034 		fl.l_type = F_RDLCK;
2035 	if (argp->block) {
2036 		struct nlm_async_lock *af;
2037 		CLIENT *client;
2038 		struct nlm_grantcookie cookie;
2039 
2040 		/*
2041 		 * First, make sure we can contact the host's NLM.
2042 		 */
2043 		client = nlm_host_get_rpc(host, TRUE);
2044 		if (!client) {
2045 			result->stat.stat = nlm4_failed;
2046 			goto out;
2047 		}
2048 
2049 		/*
2050 		 * First we need to check and see if there is an
2051 		 * existing blocked lock that matches. This could be a
2052 		 * badly behaved client or an RPC re-send. If we find
2053 		 * one, just return nlm4_blocked.
2054 		 */
2055 		mtx_lock(&host->nh_lock);
2056 		TAILQ_FOREACH(af, &host->nh_pending, af_link) {
2057 			if (af->af_fl.l_start == fl.l_start
2058 			    && af->af_fl.l_len == fl.l_len
2059 			    && af->af_fl.l_pid == fl.l_pid
2060 			    && af->af_fl.l_type == fl.l_type) {
2061 				break;
2062 			}
2063 		}
2064 		if (!af) {
2065 			cookie.ng_sysid = host->nh_sysid;
2066 			cookie.ng_cookie = host->nh_grantcookie++;
2067 		}
2068 		mtx_unlock(&host->nh_lock);
2069 		if (af) {
2070 			CLNT_RELEASE(client);
2071 			result->stat.stat = nlm4_blocked;
2072 			goto out;
2073 		}
2074 
2075 		af = malloc(sizeof(struct nlm_async_lock), M_NLM,
2076 		    M_WAITOK|M_ZERO);
2077 		TASK_INIT(&af->af_task, 0, nlm_lock_callback, af);
2078 		af->af_vp = vs.vs_vp;
2079 		af->af_fl = fl;
2080 		af->af_host = host;
2081 		af->af_rpc = client;
2082 		/*
2083 		 * We use M_RPC here so that we can xdr_free the thing
2084 		 * later.
2085 		 */
2086 		nlm_make_netobj(&af->af_granted.cookie,
2087 		    (caddr_t)&cookie, sizeof(cookie), M_RPC);
2088 		af->af_granted.exclusive = argp->exclusive;
2089 		af->af_granted.alock.caller_name =
2090 			strdup(argp->alock.caller_name, M_RPC);
2091 		nlm_copy_netobj(&af->af_granted.alock.fh,
2092 		    &argp->alock.fh, M_RPC);
2093 		nlm_copy_netobj(&af->af_granted.alock.oh,
2094 		    &argp->alock.oh, M_RPC);
2095 		af->af_granted.alock.svid = argp->alock.svid;
2096 		af->af_granted.alock.l_offset = argp->alock.l_offset;
2097 		af->af_granted.alock.l_len = argp->alock.l_len;
2098 
2099 		/*
2100 		 * Put the entry on the pending list before calling
2101 		 * VOP_ADVLOCKASYNC. We do this in case the lock
2102 		 * request was blocked (returning EINPROGRESS) but
2103 		 * then granted before we manage to run again. The
2104 		 * client may receive the granted message before we
2105 		 * send our blocked reply but thats their problem.
2106 		 */
2107 		mtx_lock(&host->nh_lock);
2108 		TAILQ_INSERT_TAIL(&host->nh_pending, af, af_link);
2109 		mtx_unlock(&host->nh_lock);
2110 
2111 		error = VOP_ADVLOCKASYNC(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE,
2112 		    &af->af_task, &af->af_cookie);
2113 
2114 		/*
2115 		 * If the lock completed synchronously, just free the
2116 		 * tracking structure now.
2117 		 */
2118 		if (error != EINPROGRESS) {
2119 			CLNT_RELEASE(af->af_rpc);
2120 			mtx_lock(&host->nh_lock);
2121 			TAILQ_REMOVE(&host->nh_pending, af, af_link);
2122 			mtx_unlock(&host->nh_lock);
2123 			xdr_free((xdrproc_t) xdr_nlm4_testargs,
2124 			    &af->af_granted);
2125 			free(af, M_NLM);
2126 		} else {
2127 			NLM_DEBUG(2, "NLM: pending async lock %p for %s "
2128 			    "(sysid %d)\n", af, host->nh_caller_name, sysid);
2129 			/*
2130 			 * Don't vrele the vnode just yet - this must
2131 			 * wait until either the async callback
2132 			 * happens or the lock is cancelled.
2133 			 */
2134 			vs.vs_vp = NULL;
2135 		}
2136 	} else {
2137 		error = VOP_ADVLOCK(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE);
2138 	}
2139 
2140 	if (error) {
2141 		if (error == EINPROGRESS) {
2142 			result->stat.stat = nlm4_blocked;
2143 		} else if (error == EDEADLK) {
2144 			result->stat.stat = nlm4_deadlck;
2145 		} else if (error == EAGAIN) {
2146 			result->stat.stat = nlm4_denied;
2147 		} else {
2148 			result->stat.stat = nlm4_failed;
2149 		}
2150 	} else {
2151 		if (monitor)
2152 			nlm_host_monitor(host, argp->state);
2153 		result->stat.stat = nlm4_granted;
2154 	}
2155 
2156 out:
2157 	nlm_release_vfs_state(&vs);
2158 	if (rpcp)
2159 		*rpcp = nlm_host_get_rpc(host, TRUE);
2160 	nlm_host_release(host);
2161 	return (0);
2162 }
2163 
2164 int
2165 nlm_do_cancel(nlm4_cancargs *argp, nlm4_res *result, struct svc_req *rqstp,
2166     CLIENT **rpcp)
2167 {
2168 	fhandle_t fh;
2169 	struct vfs_state vs;
2170 	struct nlm_host *host;
2171 	int error, sysid;
2172 	struct flock fl;
2173 	struct nlm_async_lock *af;
2174 
2175 	memset(result, 0, sizeof(*result));
2176 	memset(&vs, 0, sizeof(vs));
2177 
2178 	host = nlm_find_host_by_name(argp->alock.caller_name,
2179 	    svc_getrpccaller(rqstp), rqstp->rq_vers);
2180 	if (!host) {
2181 		result->stat.stat = nlm4_denied_nolocks;
2182 		return (ENOMEM);
2183 	}
2184 
2185 	NLM_DEBUG(3, "nlm_do_cancel(): caller_name = %s (sysid = %d)\n",
2186 	    host->nh_caller_name, host->nh_sysid);
2187 
2188 	nlm_check_expired_locks(host);
2189 	sysid = host->nh_sysid;
2190 
2191 	nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
2192 	nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2193 
2194 	if (time_uptime < nlm_grace_threshold) {
2195 		result->stat.stat = nlm4_denied_grace_period;
2196 		goto out;
2197 	}
2198 
2199 	error = nlm_get_vfs_state(host, rqstp, &fh, &vs, (accmode_t)0);
2200 	if (error) {
2201 		result->stat.stat = nlm_convert_error(error);
2202 		goto out;
2203 	}
2204 
2205 	fl.l_start = argp->alock.l_offset;
2206 	fl.l_len = argp->alock.l_len;
2207 	fl.l_pid = argp->alock.svid;
2208 	fl.l_sysid = sysid;
2209 	fl.l_whence = SEEK_SET;
2210 	if (argp->exclusive)
2211 		fl.l_type = F_WRLCK;
2212 	else
2213 		fl.l_type = F_RDLCK;
2214 
2215 	/*
2216 	 * First we need to try and find the async lock request - if
2217 	 * there isn't one, we give up and return nlm4_denied.
2218 	 */
2219 	mtx_lock(&host->nh_lock);
2220 
2221 	TAILQ_FOREACH(af, &host->nh_pending, af_link) {
2222 		if (af->af_fl.l_start == fl.l_start
2223 		    && af->af_fl.l_len == fl.l_len
2224 		    && af->af_fl.l_pid == fl.l_pid
2225 		    && af->af_fl.l_type == fl.l_type) {
2226 			break;
2227 		}
2228 	}
2229 
2230 	if (!af) {
2231 		mtx_unlock(&host->nh_lock);
2232 		result->stat.stat = nlm4_denied;
2233 		goto out;
2234 	}
2235 
2236 	error = nlm_cancel_async_lock(af);
2237 
2238 	if (error) {
2239 		result->stat.stat = nlm4_denied;
2240 	} else {
2241 		result->stat.stat = nlm4_granted;
2242 	}
2243 
2244 	mtx_unlock(&host->nh_lock);
2245 
2246 out:
2247 	nlm_release_vfs_state(&vs);
2248 	if (rpcp)
2249 		*rpcp = nlm_host_get_rpc(host, TRUE);
2250 	nlm_host_release(host);
2251 	return (0);
2252 }
2253 
2254 int
2255 nlm_do_unlock(nlm4_unlockargs *argp, nlm4_res *result, struct svc_req *rqstp,
2256     CLIENT **rpcp)
2257 {
2258 	fhandle_t fh;
2259 	struct vfs_state vs;
2260 	struct nlm_host *host;
2261 	int error, sysid;
2262 	struct flock fl;
2263 
2264 	memset(result, 0, sizeof(*result));
2265 	memset(&vs, 0, sizeof(vs));
2266 
2267 	host = nlm_find_host_by_name(argp->alock.caller_name,
2268 	    svc_getrpccaller(rqstp), rqstp->rq_vers);
2269 	if (!host) {
2270 		result->stat.stat = nlm4_denied_nolocks;
2271 		return (ENOMEM);
2272 	}
2273 
2274 	NLM_DEBUG(3, "nlm_do_unlock(): caller_name = %s (sysid = %d)\n",
2275 	    host->nh_caller_name, host->nh_sysid);
2276 
2277 	nlm_check_expired_locks(host);
2278 	sysid = host->nh_sysid;
2279 
2280 	nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
2281 	nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2282 
2283 	if (time_uptime < nlm_grace_threshold) {
2284 		result->stat.stat = nlm4_denied_grace_period;
2285 		goto out;
2286 	}
2287 
2288 	error = nlm_get_vfs_state(host, rqstp, &fh, &vs, (accmode_t)0);
2289 	if (error) {
2290 		result->stat.stat = nlm_convert_error(error);
2291 		goto out;
2292 	}
2293 
2294 	fl.l_start = argp->alock.l_offset;
2295 	fl.l_len = argp->alock.l_len;
2296 	fl.l_pid = argp->alock.svid;
2297 	fl.l_sysid = sysid;
2298 	fl.l_whence = SEEK_SET;
2299 	fl.l_type = F_UNLCK;
2300 	error = VOP_ADVLOCK(vs.vs_vp, NULL, F_UNLCK, &fl, F_REMOTE);
2301 
2302 	/*
2303 	 * Ignore the error - there is no result code for failure,
2304 	 * only for grace period.
2305 	 */
2306 	result->stat.stat = nlm4_granted;
2307 
2308 out:
2309 	nlm_release_vfs_state(&vs);
2310 	if (rpcp)
2311 		*rpcp = nlm_host_get_rpc(host, TRUE);
2312 	nlm_host_release(host);
2313 	return (0);
2314 }
2315 
2316 int
2317 nlm_do_granted(nlm4_testargs *argp, nlm4_res *result, struct svc_req *rqstp,
2318 
2319     CLIENT **rpcp)
2320 {
2321 	struct nlm_host *host;
2322 	struct nlm_waiting_lock *nw;
2323 
2324 	memset(result, 0, sizeof(*result));
2325 
2326 	host = nlm_find_host_by_addr(svc_getrpccaller(rqstp), rqstp->rq_vers);
2327 	if (!host) {
2328 		result->stat.stat = nlm4_denied_nolocks;
2329 		return (ENOMEM);
2330 	}
2331 
2332 	nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2333 	result->stat.stat = nlm4_denied;
2334 	KFAIL_POINT_CODE(DEBUG_FP, nlm_deny_grant, goto out);
2335 
2336 	mtx_lock(&nlm_global_lock);
2337 	TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
2338 		if (!nw->nw_waiting)
2339 			continue;
2340 		if (argp->alock.svid == nw->nw_lock.svid
2341 		    && argp->alock.l_offset == nw->nw_lock.l_offset
2342 		    && argp->alock.l_len == nw->nw_lock.l_len
2343 		    && argp->alock.fh.n_len == nw->nw_lock.fh.n_len
2344 		    && !memcmp(argp->alock.fh.n_bytes, nw->nw_lock.fh.n_bytes,
2345 			nw->nw_lock.fh.n_len)) {
2346 			nw->nw_waiting = FALSE;
2347 			wakeup(nw);
2348 			result->stat.stat = nlm4_granted;
2349 			break;
2350 		}
2351 	}
2352 	mtx_unlock(&nlm_global_lock);
2353 
2354 out:
2355 	if (rpcp)
2356 		*rpcp = nlm_host_get_rpc(host, TRUE);
2357 	nlm_host_release(host);
2358 	return (0);
2359 }
2360 
2361 void
2362 nlm_do_granted_res(nlm4_res *argp, struct svc_req *rqstp)
2363 {
2364 	struct nlm_host *host = NULL;
2365 	struct nlm_async_lock *af = NULL;
2366 	int error;
2367 
2368 	if (argp->cookie.n_len != sizeof(struct nlm_grantcookie)) {
2369 		NLM_DEBUG(1, "NLM: bogus grant cookie");
2370 		goto out;
2371 	}
2372 
2373 	host = nlm_find_host_by_sysid(ng_sysid(&argp->cookie));
2374 	if (!host) {
2375 		NLM_DEBUG(1, "NLM: Unknown host rejected our grant");
2376 		goto out;
2377 	}
2378 
2379 	mtx_lock(&host->nh_lock);
2380 	TAILQ_FOREACH(af, &host->nh_granted, af_link)
2381 	    if (ng_cookie(&argp->cookie) ==
2382 		ng_cookie(&af->af_granted.cookie))
2383 		    break;
2384 	if (af)
2385 		TAILQ_REMOVE(&host->nh_granted, af, af_link);
2386 	mtx_unlock(&host->nh_lock);
2387 
2388 	if (!af) {
2389 		NLM_DEBUG(1, "NLM: host %s (sysid %d) replied to our grant "
2390 		    "with unrecognized cookie %d:%d", host->nh_caller_name,
2391 		    host->nh_sysid, ng_sysid(&argp->cookie),
2392 		    ng_cookie(&argp->cookie));
2393 		goto out;
2394 	}
2395 
2396 	if (argp->stat.stat != nlm4_granted) {
2397 		af->af_fl.l_type = F_UNLCK;
2398 		error = VOP_ADVLOCK(af->af_vp, NULL, F_UNLCK, &af->af_fl, F_REMOTE);
2399 		if (error) {
2400 			NLM_DEBUG(1, "NLM: host %s (sysid %d) rejected our grant "
2401 			    "and we failed to unlock (%d)", host->nh_caller_name,
2402 			    host->nh_sysid, error);
2403 			goto out;
2404 		}
2405 
2406 		NLM_DEBUG(5, "NLM: async lock %p rejected by host %s (sysid %d)",
2407 		    af, host->nh_caller_name, host->nh_sysid);
2408 	} else {
2409 		NLM_DEBUG(5, "NLM: async lock %p accepted by host %s (sysid %d)",
2410 		    af, host->nh_caller_name, host->nh_sysid);
2411 	}
2412 
2413  out:
2414 	if (af)
2415 		nlm_free_async_lock(af);
2416 	if (host)
2417 		nlm_host_release(host);
2418 }
2419 
2420 void
2421 nlm_do_free_all(nlm4_notify *argp)
2422 {
2423 	struct nlm_host *host, *thost;
2424 
2425 	TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, thost) {
2426 		if (!strcmp(host->nh_caller_name, argp->name))
2427 			nlm_host_notify(host, argp->state);
2428 	}
2429 }
2430 
2431 /*
2432  * Kernel module glue
2433  */
2434 static int
2435 nfslockd_modevent(module_t mod, int type, void *data)
2436 {
2437 
2438 	return (0);
2439 }
2440 static moduledata_t nfslockd_mod = {
2441 	"nfslockd",
2442 	nfslockd_modevent,
2443 	NULL,
2444 };
2445 DECLARE_MODULE(nfslockd, nfslockd_mod, SI_SUB_VFS, SI_ORDER_ANY);
2446 
2447 /* So that loader and kldload(2) can find us, wherever we are.. */
2448 MODULE_DEPEND(nfslockd, krpc, 1, 1, 1);
2449 MODULE_DEPEND(nfslockd, nfslock, 1, 1, 1);
2450 MODULE_VERSION(nfslockd, 1);
2451